Space in the data centers for servers and other computer systems is often limited. To conserve space, computer components are placed in close proximity. Cases which house servers and other computer systems are designed to take up as little space as possible. In particular, rack-mountable cases used with high density server racks are typically designed to be thin and flat in order to allow multiple servers or computer systems to be installed within the same server rack. As microprocessors and electronics become more powerful, it has proven difficult to provide cooling systems which can fit within the confined area of the cases and adequately cool the devices.
Modern computers and other electronic devices usually consume large amounts of energy, most of which is converted into heat. In particular, the power requirements of small servers are usually hundreds of watts. A large percentage of this energy is used by various chips and semiconductor devices, to power hard drive motors, fans, and other electromagnetic parts, and some of this energy is spent heating wires and producing electromagnetic waves that are usually blocked by the surrounding cases and shields. Only a very small amount of this energy leaves the computer as electric signals and light. Almost all energy that is consumed by a modern computer is ultimately converted into heat.
Out of those hundreds of watts, central processing units (CPUs) and other chips dissipate a large percentage. Modern CPUs dissipate 60–90W each, and the traditional rack-mountable case designs and server chassis designs often place 2 to 4 CPUs within a flat case approximately 17 inches wide and 1.75 inches high (the depth varies between 14 inches and 29 inches). CPUs have small dies, often less than 0.5 inches by 0.5 inches, and sometimes have built-in heat spreaders up to 1 inch×1 inch. Design requirements cause those chips to be placed on the same board within a few inches from each other, thus forming a small volume where a large amount of heat is produced (120–300W depending on the models and number of CPUs). The surfaces of the heat sources are a small fraction of a square inch, so the heat production is unevenly distributed within this already small volume.
The temperatures achieved on those chips have to be within an acceptable range, usually below 90° C. for CPUs, however some CPUs and many other chips are rated to only 65° C. Reliability requirements cause hardware designers to keep CPUs under 60–70° C. and other components below 40–50° C., even under the highest load possible (where highest load means a mode of operation with the highest power consumption, which usually means the highest rate of operations and highest number of gates involved). To achieve this, the heat should be removed at the same rate as it is being produced, and the temperature equilibrium between the chip under the highest load and the cooling system should be reached below those temperatures.
Heat produced within a case surrounded by cooler moving air will eventually pass to the outside air, so while removing the heat is simple, the problem is to keep heat-producing components, including CPUs, within the temperature limits suitable for their operation at all times. The most common solution for this problem is a large air-cooled heatsink placed on the top of the CPU die, sometimes with a separate heat spreader attached to the die as part of the CPU assembly with the heat sink placed on top. Tall fins or spikes protrude from the heatsink base, and a large amount of airflow is passed along those fins to remove the heat to the outside air. Traditional designs of this type of heatsink are commonly used in computers. The most common design is a copper or aluminum plate placed on the top of the CPU (for the purpose of this description the chips are assumed to be placed on the top side of a horizontal board, which is the standard layout in rack-mountable servers), with fins attached to the plate and a flat fan above the fins. The fan creates airflow through the fins toward the base plate and outside of the device into two or more directions. The base plate transfers the heat from the CPU to the fins, and the fins transfer the heat to the moving air.
The size of the heat sink is limited due to the limited heat conductivity of the base plate and fins. The temperature is highest at the center of the heatsink and decreases toward the edges and the top of the fins. When the heatsink size is increased, there is lower average temperature difference between the moving air and the fin surface, which causes less overall efficiency. Also a large heatsink can shield components under it from the airflow and expose surrounding components to hot air exiting the heatsink causing overheating of those components. This causes engineers to limit a heatsink's footprint and increase the airflow, often including large fans, blowers, ducts and shrouds in their designs. There are also designs incorporating heat pipes to distribute heat to the edge of the base plate or to the upper ⅔ of the fins.
Heat pipes typically include a sealed vessel with a vacuum formed inside. One end of a heat pipe (the evaporator) attaches to the surface of a heat source, and the other end of the heat pipe (the condenser) extends away from the heat source and is attached to a heat exchanger, heatsink, or exposed to cooler air. Heat pipes include a wick structure inside the vessel and a working fluid inserted in the vessel to saturate the wick structure. The atmosphere in the vessel is set such that, absent heat transfer, the fluid is maintained at a liquid-vapor equilibrium.
When a heat pipe is attached to a heat source, generated heat is transferred from the heat source to the evaporator of the heat pipe. The phase change of the fluid from liquid to vapor results in the absorption of a substantial amount of heat. This transfer of heat results in the generation of higher vapor pressure at the evaporator. The vapor pressure at the evaporator causes the vapor to flow in a direction toward the condenser. The lower temperature at the condenser causes the vapor to condense back to a liquid, thereby releasing its latent heat of vaporization to the condenser. The condensed fluid saturates the wick structure and gets pumped back to the evaporator of the heat pipe by capillary forces developed in the wick structure. This continuous cycle of vaporization-condensation allows heat pipes to transfer large quantities of heat with very low thermal gradients. Air-cooled heat pipes may also include a plurality of fins around the outside surface of the condenser to enhance heat dissipation from the condenser to the surrounding air, thereby keeping the walls of the condenser cooler and increasing the heat transfer performance of the heat pipe. Heatsinks that incorporate heat pipes usually have the evaporator ends of the pipes embedded in the base plate, and the condenser ends of the same pipes either have cooling fins on the surface, or are attached to the upper section of fins that have their lower end attached to the base plate, thus distributing the heat evenly over the length of the fin.
While these designs improve the performance of the heatsinks, they increase the necessary thickness of the base plate and often require complex manufacturing procedures to ensure the efficient heat transfer to and from the heat pipes.
Flat 1.75-inch high cases that are often used for high-density servers cause additional problems for traditional heatsinks. The height of the fins is limited by the space remaining in the case above the CPU. Even if the fan is moved from above the heatsink to the side, thus leaving more space for cooling fins, the height of the fins will be less than what is commonly used in traditional heatsinks. Air ducts passing through the heatsinks can isolate the airflow and increase the efficiency, however, they have to be routed around other components, which often limits their cross-section and causes additional resistance to the airflow. As a result, larger and more powerful fans and blowers are required.
Moving the heat to a heatsink located outside of the circuit board's footprint can solve some of these problems, but traditional heat pipes have to be large enough to pass liquid over their wick to achieve the necessary efficiency. The space within a case is often limited, and CPUs are often surrounded by other tall components, in particular capacitors of their own power supply circuits and connectors for various boards and cables. Heat pipe-based cooling systems are used for small laptop computers, where the cooling requirements are less, or in larger desktop cases, where extra space is available. Heat pipe-based cooling systems small enough to fit within smaller cases, such as cases used with high density servers, do not efficiently transfer heat from CPUs with higher power requirements.
A thermosyphon is a heat pipe that lacks a wick or a wick-like structure, although a wick may be used to support evaporation of the liquid coolant. A thermosyphon relies on gravity to return condensed coolant back to the evaporator instead of the capillary effect in a wick, which relies on surface tension. Using a thermosyphon simplifies the device and allows faster coolant flow, however it has an obvious disadvantage of requiring a certain location of the evaporator and condenser relative to the direction of the force of gravity. Thermosyphons will not work if the evaporator is placed above the condenser.
Thermosyphons often have large evaporators where a nearly constant level of the liquid coolant is maintained. Evaporators have flat, smooth bottoms, or have some porous material or structures completely submerged in the liquid, to assist the evaporation. Evaporators are connected to a condenser by a pipe, or by separate pipes for the liquid coolant and the vapor (known as loop thermosyphons). Although thermosyphons are well known in the art, they are not easily adaptable for use in small scale electronics.
Thermocore International Inc. (780 Eden Road, Lancaster, Pa. 17601) provides loop thermosyphons and heat pipes for thermal control in avionics and high power electronics. However, the length of these devices ranges from approximately 0.6 meters to 2 meters (see http://www.thermacore.com/thermaloop.htm). An additional device is a square U-tube loop thermosyphon having a long U-shaped evaporator spanning multiple heat sources, and a condenser placed 15 cm to 150 cm above the evaporator (Khrustalev, Dmitry, “Loop Thermosyphons for Cooling of Electronics,” available at http://www. thermacore.com/pdfs/Thermosyphons.pdf). Khrustalev also discloses a loop thermosyphon having horizontal transport lines and a large condenser with a plurality of vertical cooling fins attached to the top of the condenser (Khrustalev, Dmitry, “Loop Thermosyphons for Cooling of Electronics”). Such devices are clearly unsuitable for the limited space associated with current rack-mountable cases, which are only approximately 1.75 inches (4.4 cm) high, unless the condenser and pipes are placed outside of the case.
Beitelmal and Chandrakant, (Jan. 2002 “Two Phase Loop: Compact Thermosyphon,” Hewlett-Packard Company publication) describe a loop thermosyphon used to cool a HP Vectra VL800 desktop computer having a 1.5 GHz Pentium-4 processor. The thermosyphon described by Beitelmal and Chandrakant has a condenser that is 8.2 cm wide, 7.5 cm high and 2.6 cm deep, and an evaporator that is 3.2 cm wide, 2.9 cm high and 3.2 cm deep. Although these dimensions are adequate for a desktop computer, the condenser alone would not fit within standard rack-mountable cases. In addition, because the thermosyphon relies on gravity to transport the coolant, the condenser is placed higher than the evaporator, thereby further increasing the height of entire thermosyphon device.
There are other various designs and experimental devices based on thermosyphons, some as simple as a single vapor cavity within a heatsink base, some as complex and elaborate as “thermal buses” and multi-stage thermosyphons. None of these designs are now in common use because of their unacceptable size, complexity and cost. The main problem is the failure of the device to fit entirely into the geometry of the standard rack-mountable case while maintaining adequate cooling abilities. The modularity requirements that allow users to mix various devices in the same rack, and the cost of space in data centers prevent large external cooling devices from being used in such environment, and the needs of companies that use large numbers of identical servers in flexibly-organized space do not create enough demand for those solutions to be viable.
What is needed is a more efficient and simple cooling system for small scale computer systems and electronics that fit within today's space requirements, especially space requirements present in typical rack-mountable server cases.